Getting Started with Claude Code: Build Your First AI Agent

From zero to a working AI agent in under 30 minutes. No SDK required.

What is Claude Code?

Claude Code is Anthropic's command-line tool that turns Claude into an agentic coding assistant. It can read and write files, run shell commands, search your codebase, and maintain context across a project. But it's more than a coding tool. With the right project structure, Claude Code becomes a runtime for AI agents.

The key insight: you don't need a framework, an SDK, or complex orchestration to build useful agents. A well-structured project directory with clear instructions is enough to create agents that do real work.

How Claude Code Works

The Core Loop

CLAUDE.md - A file in your project root that Claude Code reads automatically. It contains project instructions, agent definitions, and operating rules.
File System Access - Claude Code can read, write, and edit files in your project. This is how agents store memory, generate output, and manage state.
Tool Use - Claude Code can run shell commands, search code, and interact with your system. Agents use tools to accomplish real tasks.
Persistent Context - Between sessions, your CLAUDE.md and memory files maintain the agent's knowledge and personality.

Setting Up Your First Project

Create your project directory

Pick a folder for your project. This is where your agent will live and work.

mkdir my-ai-project
cd my-ai-project

Create your CLAUDE.md

This is the most important file. It's your agent's system prompt, operating manual, and identity all in one. Claude Code reads it automatically when you start a session in this directory.

# My Project

## Who You Are
You are Alex, a personal research assistant. You help me
research topics, summarize findings, and organize information.

## How You Work
- Be concise and direct
- Save research to the research/ directory
- Always cite sources
- If unsure, say so rather than guessing

## Communication Style
- Casual and friendly
- No corporate jargon
- Short paragraphs, bullet points preferred

## Key Directories
- research/ - saved research and summaries
- memory/ - persistent context between sessions

Create a memory directory

Memory is what makes an agent more than a chatbot. Create a directory and an initial memory file.

mkdir memory

Create memory/MEMORY.md:

# Agent Memory

## User Preferences
- (Will be filled in as we learn)

## Active Projects
- (Will be filled in as projects start)

## Key Decisions
- (Will be filled in as decisions are made)

Start Claude Code and talk to your agent

Open your terminal in the project directory and launch Claude Code:

claude

Claude Code will read your CLAUDE.md automatically. Your agent is now live. Start by telling it about a project or asking it to research something. As it works, it will read and write files in your project directory.

Creating an Agent with a System Prompt and Memory

The pattern is simple and powerful:

CLAUDE.md defines who the agent is, how it behaves, and what it has access to. This is the "personality and rules" layer.
memory/MEMORY.md stores persistent context that survives across sessions. The agent reads it at the start of each session and updates it as things change.
Agent-specific directories give the agent a workspace. A research agent saves to research/. A content agent saves to content/. A finance agent saves to data/.

Pro Tips

Keep CLAUDE.md under 200 lines. If it's getting long, move detailed instructions into separate files and reference them.
Tell the agent to update its own memory. Include an instruction like "Update memory/MEMORY.md when you learn something important about the user or the project."
Be specific about communication style. "Be concise" is better than nothing, but "Use bullet points, keep paragraphs under 3 sentences, never use corporate jargon" is better.
Define authority clearly. What can the agent decide on its own? What needs to be surfaced to you? Clear boundaries prevent both over-asking and over-acting.

What to Build Next

Once you have a basic agent working, here's where to go:

Add more agents. Create subdirectories for different agents with their own AGENT.md files. A research agent, a writing agent, a data analyst.
Add autonomous schedules. Use Python scripts with the Anthropic SDK to run agents on cron jobs. They work while you sleep.
Add tools. MCP (Model Context Protocol) servers give agents access to external services like APIs, databases, and web search.
Add coordination. A "chief of staff" agent that dispatches work to specialist agents and synthesizes results.

See how a full AI agent team works in production

Wolfepack runs an entire organization with one human and a team of AI agents. We document the whole thing.

Learn More at Wolfepack